London-based chip startup OLIX has raised $312 million in a Series B round at a $3.3 billion valuation, in what is now Europe’s largest-ever semiconductor funding round. The company, just two years old, is building Optical Tensor Processing Units (OTPUs) — chips that use light rather than electricity to run AI inference workloads.

The round attracted backing from Arm, quantitative trading firm Hudson River Trading, Netflix co-founder Reed Hastings, and the UK government’s Sovereign AI venture fund, a signal that photonic computing has moved from academic curiosity to national strategic priority.

The company’s flagship chip, the DX-1, is claimed to deliver more than 10,000 tokens per second per user for 100-billion-parameter models. First customer systems are targeted for the second half of 2027.

Why Photonic Chips Could Change the AI Cost Equation

The key technical distinction: OLIX’s chips move data using photons (particles of light) rather than electrons. This matters for AI inference because a large share of the energy and time in traditional GPU-based inference is consumed moving data around — not doing the actual computation.

Conventional AI chips also rely on High-Bandwidth Memory (HBM), an expensive, physically limited component that has become a bottleneck as AI model usage scales. OLIX’s optical architecture is designed to eliminate HBM entirely.

For businesses running AI workloads at scale, the implication is straightforward: if photonic chips deliver on their throughput promises, inference costs per token could fall significantly in the 2027-2028 window. That’s the cost businesses pay every time an AI model responds to a query, processes a document, or generates a report.

What This Means for Business

The compute market is diversifying. For the past three years, enterprise AI economics have been largely determined by GPU availability and pricing — specifically Nvidia’s H100 and B200 series. OLIX’s funding (and similar bets on other alternative architectures) signals that investors believe the GPU monopoly on AI compute is breakable. That competitive pressure benefits buyers.

Token costs will likely keep falling. If you’re planning AI deployments that depend on cost-per-query economics, the directional signal is clear: inference will get cheaper. OpenAI cut GPT-5.6 Luna’s price by 80% in late July 2026. Photonic and other alternative compute architectures add structural pressure on top of software-level price competition.

The UK is betting on sovereign AI capability. The UK government’s Sovereign AI fund participating in this round is significant for enterprises operating in regulated European markets. It signals a policy direction where non-US AI compute options may become available at commercial scale — which matters for data sovereignty and compliance.

2027 is when this gets real. OLIX is targeting first customer systems in H2 2027. This is not a product you can buy today. But for enterprises making multi-year AI infrastructure decisions now — choosing cloud providers, negotiating API contracts, planning internal deployments — it’s worth understanding that the compute landscape will look materially different in 18 months.

What You Should Do Now

The OLIX funding is a signal, not a decision point. Most businesses don’t buy chips directly — they access compute through cloud providers or AI API platforms. What this story tells you is:

Don’t lock in long-term AI infrastructure commitments at today’s prices. The compute market is in genuine flux. If you’re negotiating multi-year enterprise AI contracts, include pricing review clauses. What costs $X per million tokens today could cost significantly less by mid-2027.

Watch inference, not just training. Most enterprise AI use cases (customer queries, document analysis, reporting, voice interactions) are inference workloads — running an existing model, not training a new one. The economics of inference are what directly affect your AI operating costs. That’s where OLIX, and the broader shift in compute architecture, is focused.

The talent and tooling gap matters more than the chip gap. For most organisations, the constraint isn’t compute access — it’s knowing how to use AI effectively inside your operations. While the hardware layer evolves, the organisations building real AI capability in their teams are the ones who will be positioned to benefit when compute costs drop further.

The AI infrastructure buildout is accelerating from multiple directions at once. OLIX is one datapoint in a broader pattern: capital is flowing into every layer of the stack, from foundation models to novel compute hardware to enterprise orchestration platforms. For business leaders, the takeaway is that AI capability is going to get faster, cheaper, and more accessible — the question is whether your organisation is building the internal capacity to take advantage of it.